Forecasting the 2008 Presidential Election with the Time-for-Change Model

t first glance, the outcome the 2008 A presidential predict. For theoffirst time be election would appear to very difficult to in over 50 years, there will be no incumbent president or vice president in the race. Instead, the Republican Party, which has seen its popularity and electoral fortunes plummet since 2004, is pinning its hopes of retaining control of the White House on Arizona Senator John McCain—an individual who has frequently clashed with his own party’s leadership. And McCain’s Democratic opponent will be Illinois Senator Barack Obama, the first African American ever to receive a major-party presidential nomination. The absence of an incumbent president or vice president in the race and the unusual characteristics of the two major-party candidates have led to considerable uncertainty among political observers about the outlook for the November election. While President George W. Bush’s low approval ratings and the public’s overwhelmingly negative perceptions of the economy indicate a very difficult political climate I. Abramowitz, for Republicans, University McCain’s reputation as a maverick and Obama’s problems uniting Democratic voters behind his candidacy have led some analysts to suggest that a Democratic victory in November is far from certain. Polling data seem to support the argument that despite the unpopularity of his party, McCain has a realistic chance of keeping the White House in Republican hands. McCain has been running very close to Obama in most national polls since the conclusion of the presidential primaries. of the incumbent president, the state of the economy, and the length of time that the president’s party has controlled the White House ~Abramowitz 1988; 1996!. The assumption underlying the time-forchange model is that a presidential election is a referendum on the performance of the incumbent president. If this is true, then regardless of whether there is an incumbent in the race, how voters cast their ballots should be strongly influenced by their evaluation of the incumbent president’s performance. The evidence displayed in Figure 1 provides strong support for this hypothesis. This figure shows the trend in support for the candidate of the president’s party among voters who approved or disapproved of the incumbent president’s performance since 1972 when the American National Election Studies ~NES! began asking the presidential approval question. The data displayed in Figure 1 show that in these nine elections between 80 and 90% of those who approved of the incumbent president’s performance voted for the candidate of the president’s party while between 80% and 90% of those who disapproved of the incumbent president’s performance voted for the candidate of the opposing party. Even when the incumbent president was not on the ballot, in 1988 and 2000, the results were very similar. For example, the data displayed in Table 1 show that in the 2000 presidential election there was a very strong relationship between how voters cast their ballot in the presidential contest between Al Gore and George W. Bush and their evaluation of President Bill Clinton’s job performance: almost 80% of those who approved of Clinton’s performance voted for the Democrat, Gore, while over 90% of those who disapproved of Clinton’s performance voted for the Republican, Bush. Clearly some of the correlation between presidential approval and presidential vote is a byproduct of party identification. Both approval and vote choice are strongly influenced by party identification. However, even after controlling for party identification, there is a strong relationship between approval and vote choice. For the nine elections since the NES began asking the presidential job approval question, the partial correlation between approval and vote choice, controlling for party identification, has ranged from 0.29 to 0.58 withan average of 0.49. These results indicate that even though President Bush will not be on

by Alan Emory

The Time-for-Change Model
Despite uncertainty among political observers due to the unusual characteristics of the candidates and the tightness of the polls, research on U.S. presidential elections indicates that how voters will cast their ballots in November can be predicted accurately based on conditions that are known long before Election Day ~Fair 1978; Brody and Sigelman 1983; Lewis-Beck and Rice 1992; Holbrook 1996; Wlezien and Erikson 1996; Norpoth 1996; Campbell and Garand 2000; Campbell 2000; Hibbs 2000!. According to the time-for-change model, these conditions include the popularity
PSOnline www.apsanet.org

doi:10.1017/S1049096508081249

691

By itself. but that economic conditions are only one of the factors that influence voters’ evaluations of the incumbent president’s performance. Figure 2 Vote for Incumbent Party by Annual GDP Growth Rate during Second Quarter of Election Year. such as 1972 and 1984. how voters cast their ballots in November will be strongly influenced by their evaluation of his performance. the time-for-change model includes the incumbent president’s approval rating in the Gallup Poll at mid-year as another predictor of voter decision making. the candidate of the incumbent party does much better than one would have expected based on the president’s approval PS October 2008 692 . and their domestic policy agenda. The data displayed in Figure 3 show that there is a fairly strong relationship between the incumbent president’s approval rating at mid-year and the outcome of the presidential election. explains almost two-thirds of the variance in the incumbent party’s share of the major-party vote. Moreover. Since World War II there have been several elections in which the candidate of the incumbent party has done much better than one would expect based on economic conditions.S. other economic indicators such as unemployment. change in real disposable income. 1972–2004 Table 1 2000 Presidential Vote by Clinton Job Approval Vote Gore Bush Total (n) Approve 79% 21 100 (3. presidential elections.1 Presidential approval. the ballot in 2008. the growth rate of the economy explains less than 40% of the variance in the outcomes of U.471) Source: 2000 National Exit Poll. the incumbent president’s approval rating at mid-year. their honesty and integrity. 1948–2004 Source: American National Election Studies. However. their personal style and communication skills. The time-for-change model is based on three leading indicators of presidential vote choice: the growth rate of the economy during the second quarter of the election year. Nevertheless. provides a measure of national economic conditions. as measured by real annualized GDP growth. On the other hand. for example.456) Disapprove 9% 91 100 (2. Presidents are also judged on the basis of their conduct of foreign affairs. by itself. lost a close contest to Republican Richard Nixon despite a booming economy. In order to capture some of the other factors that go into the public’s evaluation of the president. and the length of time the incumbent president’s party has controlled the White House—the time-for-change factor. who chose not to run for reelection. Republican incumbent Dwight Eisenhower was reelected in a landslide despite anemic economic growth. In 1968. The lesson one should draw from Figure 2 is not that economic conditions don’t matter. The first of these indicators. Rational voters are expected to reward the party in power for good economic conditions and punish the party in power for bad economic conditions.Figure 1 Percentage Voting for Candidate of President’s Party by Approval of President’s Job Performance.S. the growth rate of the economy during the second quarter of the election year. In 1956. It is widely assumed that economic conditions have a strong influence on voting behavior in a democracy. and job creation do no better. the fit between presidential approval and election results is far from perfect. presidential elections since World War II is not very strong. the data displayed in Figure 2 indicate that the relationship between economic conditions and the outcomes of U. seeking to succeed Lyndon Johnson Source: Bureau of Economic Analysis and data compiled by author. of course. Democrat Hubert Humphrey. for example. there have been several elections in which the incumbent party has done much worse than one would expect based on economic conditions. In some years.

it does not perform nearly as well. ala Dependent Variable: vote. date of the incumbent party won the national PSOnline www.020 . Model R R Square R Square the Estimate The evidence presented in Table 2 provides a 1 . It is based on the hypothesis that voters attach a positive value to periodic alternation in power by the two major parties and Table 3 that regardless of the state of the economy and the popularity of the current president.026 party has lost six of the eight postwar presidenterm −5. the candiSource: Bureau of Economic Analysis and data compiled by author.182 . Sig. The time-for-change variable is not equivalent to incumbency. In fact. 1948–2004 Type of Election Results Won Lost Average Vote First-Term 6 1 55.733 .448 −2. when a Results of Regression Analysis of Presidential Election party has held the White House for two or Results. Jimmy Carter in 1980 was the only first-term incumbent in the past century to lose a 1 (Constant 52. 1948–2004 Table 2 Success of Incumbent Party Candidate in Presidential Elections by Type of Election. such as 1960 and 2000.826 . in the case of Al Gore in 2000.or Later-Term 2 6 49. Since 1900.321 2. voters will be more likely to feel Model Summary that it is time to give the opposing party an opportunity to govern than when a party has Adjusted Standard Error of held the White House for only one term. though most of these elections were very close and. The third predictor used in the time-for-change model is a dummy variable that measures the difference between elections in which a party has controlled the White House for one term and those in which a party has controlled the White House for two or more terms.597 tial elections in which his party had controlled inc .7973 empirical support for the time-for-change hypothesis. the president’s gdp .794 693 .618 3. When a dummy variable measuring the presence of absence of an incumbent is substituted for the time-for-change dummy variable.or later-term elections since World War II have involved an incumbent running for reelection: 1948 and 1992. Since the end of World War II. the candidate of the incumbent party does much worse than one would have expected based on the president’s approval rating at mid-year. inc. have been seven presidential elections in which a party has controlled the White House for one Coefficients a term and the president’s party has won six of Unstandardized Standardized these elections.org popular vote while losing the electoral vote. and 1992. there a Predictors: (Constant). . its impact is negligible and the explanatory power of the model is unchanged.or later-term elections in this series have involved an incumbent running for reelection: 1932. only two of the nine second.Figure 3 Vote for Incumbent Party by Incumbent President’s Approval Rating at Mid-Year. When an incumbency dummy variable is included in the model along with the time-for-change dummy variable. rating at mid-year. term. 1932–2004 more terms.598 4. Clearly there were other factors influencing voter decision making in these elections. I therefore conducted an additional test of the time-for-change hypothesis using data on 18 presidential elections between 1932 and 2004.162 presidential election.092 −.432 2.881 18. since 1900 the Coefficients Coefficients president’s party has won 10 out of 11 first-term Model B Standard Error Beta t elections. 1948. the president’s party has lost seven out of 13 elections in which his party had controlled the White House for two or more terms.045 . 1940.613 2.apsanet. This variable is intended to capture the strength of time-for-change sentiment in the electorate.000 . This allows us to distinguish more clearly between the effects of the time-for-change and incumbency variables since five of the 12 second.001 .266 the White House for two or more terms. Source: Gallup Poll and data compiled by author. Note: Vote share based on major-party vote.306 .9% Second. In other years.682 . In contrast.5% Source: Data compiled by author. 1944. However. gdp.

262 . Moreover.6%. 1980. 1988. 1956. also produces very accurate forecasts of the outcomes of the last four presidential elections Source: Data compiled by author. exTable 5 shows the estimated coefficients for the 2008 timeplaining over 60% of the variance in the popular vote.3 53.212 . and Estimation of 2008 Time-for-Change Model the time-for-change dummy variable—performs fairly well.898 1.6 50.6 51. 694 PS October 2008 . According to these results.000 In addition to explaining a large proportion q2gdp .835 61. 1984. based on data for presidential elections beimportantly.110 . and 2000!.000 of the variance in the outcomes of postwar term −4. For the four elections prior to 1948.7 Vote 46.604 . 1996.394 −4. the difference between having to explain a much larger proportion of the variance in the outcomes of postwar presidential elections than we can with a model based on any one or two of these variables. These forecasts were generated by re-estimating the model for each election based on data from earlier elections and using the preliminary estimate of real GDP growth released by the Bureau of Economic Analysis in August. For the 10 postwar elections with a running incumbent a Predictors: (Constant). 1992. effect. More for-change model. Note: Forecast and vote represent incumbent party’s share of major-party popular vote.0 +1. the results show that while the time-for-change tween 1948 and 2004. For Unstandardized Standardized the five postwar elections with no running inCoefficients Coefficients cumbent ~1952. based on information available several months before Election Day.2 Error −0.920 . 1972.474 5.Figure 4 Incumbent Party Vote by Predicted Incumbent Party Vote.1% of the major-party vote. 1964.4! has a strong and statistically significant percentage point increase in the president’s net approval rating. Thus.602 . 1976. the time-for-change model. q2gdp.001 presidential elections.013 −. incumbency. The re-estimated models predicted the winner of the popular vote in each of these elections with an average absolute error of only 1.6! is very small and statistically insignificant. The average absolute error of such pre-election polls is about 2–3 percentage points.2%. for every 1 dummy variable ~ 5.109 .1 percentage points.6% 54.417 .5 Source: Data compiled by author. of just over 0. and 2004!. 1948–2004 Table 4 Accuracy of Time-for-Change Model Forecasts Based on Prior Elections.7% 54.265 1. However. the evidence displayed in juneapp .115 .7820 tions with a running incumbent.1 +2.021 . no presidential approval data and only annual GDP data are available. was more accurate than public opinion polls conducted in the final week before each of these elections.5. While this may Combining our three predictors into a single model allows us seem like a fairly small effect.9% 0. based on data available prior to those elections. the model is almost as accurate at predicting the Adjusted Standard Error outcomes of elections without a running incumModel R R Square R Square of the Estimate bent as it is at predicting the outcomes of elec1 . Model B Standard Error Beta t Sig. term ~1948. Source: Data compiled by author. The data displayed in Figure 4 show that our three-variable time-for-change model explains 92% of the variance in the inTable 5 cumbent party’s share of the major-party vote. juneapp.2 51. the estimated coefficient for the incumbency dummy the candidate of the president’s party can expect an increase variable ~0. 1960.959 a . the average absolute error of the model’s predictions is 1. 1992–2004 Election 1992 1996 2000 2004 Forecast 45.000 Table 4 shows that the time-for-change model a Dependent Variable: vote.500 . 1968. the results displayed in Table 3 show that a model including three independent variables—annual GDP change. the average absolute Coefficients a error of the model’s predictions is 1. 2008 Forecasting Model Almost all of the predicted results are very Model Summary close to the regression line. 1 (Constant 51.

and Political Campaigns.apsanet. 2008 is a second-term election. Ray C.S. Alan I.C. according to the Gallup Poll conducted closest to mid-year ~July 10–13!. The 2008 Forecast Given these results.6% of the vote.. 1996. For every additional 1 percentage point of real annual GDP growth during the second quarter. Jr. the time-forchange model predicts that Barack Obama will receive 54. 45. Wlezien. 1978. Thousand Oaks. Note 1..3% of the major-party vote in November vs. Garand. 2000. A candidate from the president’s party running in a year when real GDP grows at an annual rate of 5 percentage points during the second quarter can expect to receive an additional 3% of the vote compared with a candidate from the president’s party running in a year when there is no growth in real GDP during the second quarter. real GDP grew at an annual rate of 3. 61% disapproval!. Thomas M. “Bread and Peace Voting in U. 2000. Hibbs. the combination of an unpopular Republican incumbent in the White House. and Lee Sigelman. Washington. “Polls and Votes: The Trial-Heat Presidential Election Forecasting Model. a weak economy. 1983. 1996. and a second-term election make a Democratic victory in November all but certain. CA: Sage Publications. President Bush’s net approval rating stood at a lowly 30% ~31% approval vs. Fair. Norpoth. Thousand Oaks. Garand. there is a dramatic difference between first-term and later-term elections. or the difference between approval and disapproval.a popular and having an unpopular president in the White House is substantial. and the length of time that the president’s party has controlled the White House. the difference between having a president with a net approval rating of 20 and having a president with a net approval rating of 20 would be 4% of the vote. the difference between running with a stagnant economy and running with a booming economy is substantial. References Abramowitz. Presidential Elections. “An Improved Model for Predicting the Outcomes of Presidential Elections. Christopher.” American Politics Quarterly 24: 434– 42.or later-term election suffers a penalty of more than 4 percentage points compared with a candidate running in a first-term election. substituting the percentage of Americans approving of the president’s performance in the model makes very little difference in the results.” American Politics Quarterly 24: 443– 67. the candidate of the president’s party can expect to receive an additional 0. James E. Thus. Do Campaigns Matter? Thousand Oaks. Before the Vote: Forecasting American National Elections. Lewis-Beck. the condition of the economy. even after controlling for the popularity of the incumbent president and the state of the economy. and. Michael S. Presidential approval is measured by net approval. 1992.7% for John McCain. 1996. Campbell. eds. 2008 appears to be the most unpredictable presidential election in more than 50 years.” Review of Economics and Statistics 60: 159–72.: Congressional Quarterly Press. and Tom W. D. Based on these results. _. Campbell and James C. “Of Time and Candidates: A Forecast for 1996.” In Before the Vote: Forecasting American National Elections. what is the outlook for the 2008 presidential election? According to the August estimate from the Bureau of Economic Analysis. This approach is less sensitive to the increase over time in the proportion of respondents expressing an opinion on the Gallup Poll’s presidential approval question.” Public Choice 104: 149–80. The results in Table 5 indicate that economic conditions have a substantial impact on the outcomes of presidential elections. then. Erikson. However. experience indicates that the outcome of the 2008 presidential election can be predicted with a high degree of accuracy based on the same fundamental forces that affect the outcomes of all presidential elections: the popularity of the incumbent president. Certainty. and James C. James E. 1996. Helmut. Richard. According to these results.org 695 . ed. “Presidential Popularity and Presidential Elections: An Update and Extension. 2000. Thus. Nevertheless.3% during the second quarter of 2008.. “Bill and Al’s Excellent Adventure: Forecasting the 1996 Presidential Election. “Temporal Horizons and Presidential Election Forecasts. the impact of the time-for-change variable is equivalent to a difference of about 40 points in net approval and about 7 points in real GDP growth.” American Politics Quarterly 24: 492–505. Brody. and the first African American presidential candidate in American history. While factors outside of the model such as rising partisan polarization and resistance to an African American candidate by some white voters may result in a somewhat smaller popular vote margin for the Democratic nominee. CA: Sage Publications. it is simply much more difficult for the president’s party to retain its hold on the White House after two or more terms in office. Campbell. Holbrook. Regardless of the popularity of the president or the state of the economy. A candidate from the president’s party running in a second. “The Effect of Economic Events on Votes for President. Rice. 1988. PSOnline www. of course.” Public Opinion Quarterly 47: 325–8. Douglas A. and Robert S. Forecasting Elections. For example.” PS: Political Science and Politics 21: 843–7. With no incumbent president or vice president in the race. a party maverick leading the Republican ticket. James E. Perhaps the most interesting finding in Table 5 involves the impact of the time-for-change variable. CA: Sage Publications.

Sign up to vote on this title
UsefulNot useful